Data-based intervention approach for Complexity-Causality measure.

Area of Science:

• Complex systems analysis
• Data-driven scientific discovery
• Causality inference

Background:

• Model-free causality estimation methods are widely used but often assume separability of cause and effect in measurements.
• Existing methods fail when cause and effect are inherently inseparable or become so in acquired data.
• These methods rely on associations between well-separated samples, limiting their applicability to real-world processes.

Purpose of the Study:

• To propose a novel causality measure capable of capturing non-associational causality.
• To address the limitations of existing model-free methods in scenarios with inseparable cause-effect relationships.
• To develop a robust and effective tool for causality testing in complex systems.

Main Methods:

• Introduced Compression-Complexity Causality, a novel measure utilizing an adaptive interventional scheme.
• Characterized complexities within dynamical process evolution over short measurement windows.
• Rigorously tested the measure on simulated and real datasets, comparing it with Granger Causality and Transfer Entropy.

Main Results:

• The proposed measure effectively captures causality even when cause and effect are inseparable.
• Compression-Complexity Causality demonstrated robustness against noise, memory, filtering, low temporal resolution, non-uniform sampling, finite signal length, and common driving variables.
• Outperformed existing state-of-the-art causality measures in rigorous testing.

Conclusions:

• Compression-Complexity Causality offers a significant advancement in causality testing, particularly for complex, real-world systems.
• The adaptive interventional scheme provides a powerful approach to uncovering non-associational causal links.
• The measure's robustness and superior performance establish it as a valuable tool for diverse scientific applications.